Social media platforms have become a significant source of real-time human expression, offering opportunities to detect early signs of mental health issues through automated analysis. In recent years, data mining techniques—including natural language processing (NLP), sentiment analysis, machine learning, and deep learning—have been widely applied to analyze users' social media content for mental health screening. Public datasets like the Reddit Self-Reported Depression Diagnosis (RSDD) and shared tasks such as CLPsych and eRisk have fueled progress, with studies reporting accuracies exceeding 90% on depression detection tasks. Transformer-based models, particularly BERT variants, have shown impressive performance, achieving over 98% accuracy in some studies. However, challenges persist. Social media data are often biased toward English-speaking, Twitter-centric users, and labeling is noisy. Models also struggle with generalization across platforms and demographics. Ethical concerns around privacy, consent, and misclassification remain critical. This paper presents a comprehensive review of the latest developments (2020–2025) in mining social media for early indicators of mental health conditions. We examine methodologies used in feature extraction, modeling, and multimodal fusion, and compare model performance using metrics such as accuracy, recall, and F1-score. The study also highlights widely used toolkits and datasets, and surveys emerging trends like large language models and multimodal analysis. Our aim is to provide a concise overview of current capabilities and challenges in the field and outline future directions for responsible and effective mental health prediction using social media data.
Gaonkar et al. (Sat,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: